First Principles Thinking

A first principle is a basic proposition or assumption that cannot be deduced from any other proposition or assumption — Aristotle’s “first basis from which a thing is known.” First principles thinking is the discipline of decomposing a problem down to those irreducibles and reasoning up from them, instead of reasoning by analogy (“it’s done this way because it’s done this way”). Both sources captured: clippings/fs-blog-first-principles.md (Farnam Street) · clippings/jamesclear-com-first-principles.md (James Clear).

The key calibration — you rarely need the atomic level: “you don’t have to simplify every problem down to the atomic level… You just need to go one or two levels deeper than most people. Different solutions present themselves at different layers of abstraction.” The Lego house: at layer one you rearrange blocks and slightly improve the house; one layer deeper you own the blocks themselves and can build something entirely different. First principles are also context-relative and revisable — the laws of thermodynamics are irreducible to a refrigerator engineer but not to a theoretical physicist; they’re a foundation, not an eternal checklist.

The two canonical techniques (fs.blog):

• Socratic questioning — systematic, not adversarial: clarify origins of your thinking → challenge assumptions → demand evidence → consider alternative perspectives → trace consequences → question the questions themselves
• Five Whys — chain “why” until the answer is something you’d defend as physics (or a decision you own), not a convention

The exhibits that made the concept famous:

• Musk’s rocket arithmetic (Clear): rockets cost $65M; the aerospace-grade aluminum/titanium/copper/carbon fiber on the commodity market cost ~2% of the price. The gap wasn’t physics, it was industry convention — SpaceX collapsed it and cut launch prices ~10×. The method: “reason from first principles rather than by analogy.”
• Boyd’s snowmobile (Clear): decompose motorboat + tank + bicycle into parts, recombine handlebars+seat+treads+motor+skis → a snowmobile. Destruction of the original categories is the creative act — Boyd’s OODA-loop lineage.
• Descartes’ methodical doubt — the philosophical extreme: doubt everything doubtable until only indubitable truths remain.
• Gutenberg: press = wine screw press + movable metal type + ink + paper (decomposed existing tools, recombined).

Where the vault already runs on this

• Musk engineering process — the five-step algorithm starts here: “make the requirement less dumb” then “delete” before optimizing; the Starbase NTO/MMH example is first principles applied to a fuel choice. Note the honest limit: steps 3–5 (simplify/optimize/automate) exist precisely because pure first-principles rebuilds are expensive and error-prone — Musk’s own Model 3 fiberglass confession.
• Performance engineering — Gregg’s USE method and the elimination ladder are first principles for latency: measure what’s irreducibly true (syscalls, cache lines) before believing folklore.
• KISS / the KRZ script convention — stripping a script to bare logic + timestamped logging is decomposition to the irreducible task; monitoring/stamping/logging get added back only as justified requirements.
• The engineering-vs-emulation split in Rewriting Ansible roles from K8s operators: taking the operator’s specification (the principle) rather than its implementation surface (the analogy).

The trap both sources flag: reasoning by analogy is cheaper — it’s how expertise normally compounds — so first principles thinking is a deliberate expense. Spend it where the environment changed (conventions went stale), where the stakes are high, or where everyone else’s answer is a copy of a copy. fs.blog’s framing: “everything that is not a law of nature is just a shared belief” — money, borders, org charts, “we’ve always done it this way.”

Related: Musk engineering process · Performance engineering · Working with legacy code (fear is inherited, not deduced) · AI engineering